If the baseline does not tie out, nothing downstream matters

Every network design engagement has a moment where someone senior asks why the model says freight costs a number that does not match what finance reports. If that question comes up during the final readout, the study is finished regardless of how good the optimization was.

Baseline reconciliation is the phase that prevents this. It is unglamorous, it consumes a disproportionate share of the timeline, and it is the single strongest predictor of whether the recommendation gets implemented.

The model, run against last year of actual demand with the actual footprint and actual policies, should reproduce last year of actual cost within a few percent by category. Not in total, by category. A total that matches while transportation is high and warehousing is low is two errors cancelling, and they will not cancel in the scenarios.

Getting there surfaces problems that matter well beyond the study: lanes billed under the wrong facility, product weights that were never maintained, accessorial charges nobody could allocate, and customers served from a site the master data says does not serve them.

Schedule pressure, mostly. The reconciliation phase produces no exciting output, and there is always a temptation to move to scenarios where the interesting conversation is. Every study that skips it arrives at the same place eventually, just later and with less credibility remaining.

The other reason is that reconciliation surfaces data quality problems that are owned by other teams. That is a political cost paid early rather than an analytical cost paid late, and it is worth paying.